Memory

[prashanth]

Dear All ,

We have 32 GB Physical Ram in our server and two device swap .

             Mb      Mb      Mb   PCT  START/      Mb
TYPE      AVAIL    USED    FREE  USED   LIMIT RESERVE  PRI  NAME
dev       16384       0   16384    0%       0       -    1  /dev/vg00/swap
dev        8192       0    8192    0%       0       -    2  /dev/vg00/swap2
reserve       -     357    -357
memory    25457    4508   20949   18%
total     50033    4865   45168   10%       -       0    -

Now We are planning to increase the physical memory by 32GB so total memory will be 64GB . I have couple of questions on swap
1) Is 24 GB sufficient swap for 64 GB
2) Can we go ahead on san disk base swap space? if yes what are the advantages and disadvantages
3) If I am going with san disk based swap space which will be good device swap or file system swap
4) Which will be good if I am assigning the one lun for swap space creation then which approach will be good whole disk approach or swap space on logical volume approach?
5) do I need to change any kernel parameter for it.

Thanks!!

[Bill Hassell]

The short answer is that you don't want to swap anything if you want performance. Swapping anything will really slow down your system. So it really doesn't matter where you put your secondary swap space. There are no advantages to internal or JBOD versus SAN-based swap. Some will say that SAN swap is more vulnerable because it can be taken away by a bad SAN admin (which crashes the system). On the other hand, if the SAN storage is taken away, most likely the system is useless anyway.

You need swap space to cover the amount of active process space, but not for buffer cache and kernel areas. With only 24GB of swap space, you will probably hit a maping limit depending on the version of HP-UX you are using. I would map about 48GB total swap space so get 24GB from the SAN and you should be OK.

[Dennis Handly]

>1) Is 24 GB sufficient swap for 64 GB

Probably.  If you have that much RAM, you don't need much more swap.

[prashanth]

Thanks you for support . It will be good if you through some light on below points

4) Which will be good if I am assigning the one lun for swap space creation then which approach will be good whole disk approach or swap space on logical volume approach?
5) do I need to change any kernel parameter for it.

Thanks in Advance!!

[aaee]

Hi,

Sorry to hijack you thread, as we are doing very much the same thing but from approx 48 to >90gig. But the response may help you as well.

We will probably leave the paging at around 50gig, as it does not page (most days) but:

Dump space, does that need to be increased as well ? I know it needs to be contiguous so we may need to add an additional, rather than expand ? Is that recommended as we don't really want a 90gig dump ..............

Ulimit, coredump(blocks)  is already set to 4194303.

Thanks

[Dennis Handly]

>Dump space, does that need to be increased as well?

Perhaps but if you need that much, something is terribly wrong.

>Is that recommended as we don't really want a 90gig dump

Right.  /sbin/crashconf should show you what's being dumped.  If it isn't dumping user pages, that should limit it.

>ulimit, coredump(blocks)  is already set to 4194303.

This is different than panic dumps.

[Bill Hassell]

> Dump space, does that need to be increased as well...

The tool to fine tune crash dump space (not core files from failed processes) is called crashconf (11.11 and later), but the reality is that 90% of crash dumps only need a few GB for analysis. The kernel and driver areas are the important parts. User process area and buffer cache are not needed much. I have analyzed crash dumps from 128GB systems that had only 6GB of crash space. If you make your dump space the same as swap, you'll be fine. However, you'll very likely need to create a new lvol for /var/crash, about 25% the size of your dump area.

ulimit has nothing to do with HP-UX system crashes. ulimit -a will report on the maximum size of a program's core file, but unless your system is being used by programmers that write bad code, or you are an unwilling tester of some vendor's code, you can eliminate all core dumps completely by adding this to /etc/profile:

ulimit -Sc 0

Now there will be no core dumps from any user that logs in and has program crashes. If a programmer wants the core file, they can raise the value from 0 (the -S is for a soft setting).

[Adrian Hicks]

This maybe helpfull for sizing "dump"

After you adding the memeory , you can determine the dump size required on the server by below equation.

 Dump-space-needed = 2.5 GB + (1/10 * system RAM).

#crashconf –v

Enjoy :)

[Torsten.]

The memory - swap/dump relation is always a difficult topic.

Since an hp-ux server can currently have up to 4 TB RAM, you will always decide to have less swap/dump space than real memory. The values may depend on your configuration and application needs.